Research On Intelligent Fault Diagnosis Of Power Transformer Based On Dissolved Gas Analysis In Oil

Power transformer plays an essential role in the process of power transmission and distribution in the power system.Its operation status directly influences the safety and reliability of current supply in the power system.Exact identification of potential faults of power transformers,reduction of economic losses and hazards caused by faults have long been hot research directions in the field of electrical engineering.The most widely used method in transformer fault diagnosis is dissolved gas analysis,but the traditional fault identification method based on dissolved gas in oil has low diagnostic accuracy,and has the limitations of too absolute boundary and incomplete coding.Based on the analysis of existing transformer fault diagnosis methods,this paper combines traditional methods and artificial intelligence methods to achieve transformer fault diagnosis,which has key theoretical and practical meaning for the secure and steady operation of power system.The main work is as follows:(1)Summarize the mapping relationship between transformer fault and its produced features gas,and combined with traditional methods such as non-coding ratio,the dimension of the collected transformer oil and gas features data is reduced through data dimension reduction,reduce the redundant features of the model input and improve the diagnosis speed of the model,and obtain the oil and gas features that are more relevant to the transformer faults.(2)Aiming at the problems of low accuracy of existing fault diagnosis methods and deficiency generalization performance of models,transformer fault diagnosis models based on regularized extreme learning machine and kernel extreme learning machine are established,the over-fitting problem of model training is improved.The experimental results show that the fault diagnosis accuracy has been further promoted.(3)Owing to the hyperparameters of the constructed classification model have a significant affect on the experimental results,for this reason,the Capuchin Search Algorithm and the Firefly Algorithm improved Salp Swarm Algorithm are introduced into the model to optimize the parameters of the input layer weight and the hidden layer bias of the extreme learning machine,effectively avoiding the problem of the model sinking into the local optimal solution,and improving the stability of the model.(4)Compared with the traditional transformer fault diagnosis methods such as BP neural network,Probabilistic neural network,Extreme learning machine,and the traditional optimization algorithm such as genetic algorithm to optimize the extreme learning machine,experiments are carried out.The experimental results show that the overall performance of Cap SA-RELM and FASSA-KELM is better than the traditional classification model,the fault diagnosis accuracy is improved,the generalization and stability of the model are improved,and the floating error of the fault diagnosis accuracy is small.